Temporal Pattern Analysis of Baggage Impact on Flight Operations

Necip Gozuacik

, Adem Tekinbas

, Engin Sag

, Onur Adiguzel

and Sibel Malkos

Technology and Innovation Department, Siemens Kartal R&D Center, Istanbul, Turkey

Keywords: Baggage Analytics, Big Data, Exploratory Data Analysis, Pattern Analysis, Time-Based Baggage Records.

Abstract: Transportation hubs like airports are increasingly complex due to globalization, resulting in diverse operations

and stakeholders with often conflicting objectives. Operational inefficiencies, exacerbated by unpredictable

demand and external factors such as weather, can cause economic losses and reduce sustainability. This paper

aims to apply holistic approach via exploratory data analysis, integration of datasets, advanced pre-processing

techniques and reveal pattern analysis. This will be a prerequisite work for solving the inefficiencies by

centralizing monitoring and tracking of all operations Additionally, given the large volume of structured and

unstructured data, this study underscores the importance of big data processing. Utilizing NoSQL database

technology, specifically Cassandra, enables scalable, high-performance handling of millions of rows of data.

The conclusions offer insights about the importance of temporal pattern analysis regarding future AI

developments.

1 INTRODUCTION

Transportation hubs, such as airports, are facing

heightened complexity due to globalization, with an

increasing variety of operations and stakeholders

involved in the process chain. These stakeholders

may have competing business objectives or internal

conflicts (SOCFAI, 2024).

Operational inefficiencies are further intensified

by unpredictable demand and external factors like

weather, resulting in economic losses and reduced

sustainability (Anupkumar, 2023). Measuring

operational delays is essential to identify the

inefficiencies, but integrating systems across

operations and hubs is an overly complex process that

often necessitates custom solutions. Enhancing

efficiency requires centralized monitoring and

tracking of all operations.

In this work, pattern analysis addressing a holistic

approach is applied to flight and baggage datasets

collected from several airports. This approach utilizes

advanced analytical techniques to identify patterns,

relationships, and anomalies within the datasets,

https://orcid.org/0000-0003-0261-4404

https://orcid.org/0000-0001-7935-7309

https://orcid.org/0009-0001-7476-9803

https://orcid.org/0000-0002-0195-4311

https://orcid.org/0000-0002-2159-5766

ensuring a comprehensive understanding of the

operational dynamics. By integrating domain

knowledge with data-driven insights, the analysis

aims to discover latent connections between baggage

handling and flight operations, providing actionable

intelligence for optimizing processes. Additionally,

the holistic perspective ensures that the interplay of

numerous factors, such as airline schedules and

passenger behaviors, is thoroughly considered. The

findings from this analysis can support decision-

making processes, improve operational efficiency,

and contribute to the development of predictive

models for future scenarios.

Original research contribution and novelty here

are to perform temporal pattern analysis of baggage

records and reveal correlation between baggage

analytics and flight operations with considering

holistic approach. To achieve that analysis, various

steps are applied such as data cleaning, irrelevant data

removal, joint dataset generation linking flight and

baggage datasets, feature extraction etc.

This paper consists of five sections. Section 2

gives information about activities and insights in

literature. In Section 3, exploratory data analysis is

Gozuacik, N., Tekinbas, A., Sag, E., Adiguzel, O. and Malkos, S.

Temporal Pattern Analysis of Baggage Impact on Flight Operations.

DOI: 10.5220/0013372800003905

In Proceedings of the 14th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2025), pages 831-838

ISBN: 978-989-758-730-6; ISSN: 2184-4313

831

studied and outlined. Pattern analysis and interpreted

results are included in Section 4. Finally, Section 5

concludes the paper.

2 LITERATURE REVIEW

In literature, there are several studies regarding

applying pattern analysis and AI technology for

airport and aircraft operations. From the overall

perspective, majority of the publications focus on

flight and passenger aspects. In contrast to this, there

are less studies interest in baggage operations and

their effects on airport operations and as well as flight

operations.

In a study, existing research on AI applications in

airports are synthesized and future research directions

are identified (Amiri and Kusakci, 2024). The

findings indicate a significant focus on airport

administration and management, with AI enhancing

security, traffic management, and passenger

experience.

Developing predictive models for passenger

arrival patterns at airport counters are studied

specifically at İzmir Adnan Menderes Airport's

international terminal, to enhance operational

efficiency (Sayın et al., 2023). The research compares

various predictive models, including linear regression

and LSTM. This study contributes to the literature by

providing insights into effective predictive modelling

for airport operations, which can lead to improved

resource allocation and passenger satisfaction

Another paper aims to improve the prediction of

checked baggage volumes for flights, which is crucial

for efficient aircraft load planning (Chen et al., 2024).

The study utilizes a dataset from a major US airline,

applying three machine learning algorithms.

Another study aims to provide a comprehensive

review of demand analysis and forecasting algorithms

for checked baggage flow at airports, addressing the

inefficiencies in baggage handling (Jiang et al.,

2024). The research employs a mixed-method

approach, integrating quantitative data analysis with

theoretical modelling.

A thesis study investigates to enhance the

accuracy of baggage loading and un-loading duration

predictions during KLM's turnaround process using

data-driven and machine learning methods (Ochoa

Barnuevo, 2023). The study contributes to both

theoretical advancements in machine learning

applications in aviation and practical implications for

KLM's operational efficiency and customer

satisfaction.

Real-time data sharing and collaborative decision-

making process, particularly for transfer passengers

are important to enhance operational efficiency (Guo

et al., 2020). The mentioning study contributes

theoretically by advancing the understanding of

predictive analytics in airport operations and by

providing a model that can be adopted by other

airports to optimize their operations.

A standard set of measures is proposed to assess

the expected performance of a baggage handling

system through discrete event simulation (Le et al.,

2012). A stochastic, periodic input modelling

methodology from real and simulated data sources is

proposed.

Dimensionality reduction methodology is also

important especially for large volumes of data to

enhance performance of pattern analysis and

recognition (Petersen et al., 2024).

The state-of-the-art in the current approaches are

often limited to a single perspective, focusing

exclusively on specific domains such as flight-

centric, passenger-centric, or baggage-centric views.

In our study, the original contribution here is to

evaluate baggage dataset from several aspects and

outline various pattern analysis with combining flight

dataset as a holistic view.

The advantage and novelty of this paper are to

perform deep-dive analysis about time-based

baggage records and reveal their impact on flight

operations from the point of pattern analysis view.

The highlighted points are going to be used as new

feature candidates for predicting baggage operation

volume and impacted flight operations in real world

scenarios.

This will be a baseline study from the point

of data pre-processing and feature extraction for

further AI model developments on this subject.

3 EXPLORATORY DATA

ANALYSIS

In this section, provided dataset regarding the study is

explored prior to pattern analysis. Content of the

dataset is investigated from the point of several

aspects such as descriptive analysis, pre-processing

operation, storage strategy and visualization.

3.1 Dataset Description

Two datasets regarding flight and baggage

information are analyzed in the study. The datasets

are collected from several airports in Turkiye as part

of the SOCFAI project and have been shared in this

ICPRAM 2025 - 14th International Conference on Pattern Recognition Applications and Methods

832

publication in anonymized form. The authors do not

have permission to share data. Initially, the data is

collected for one month period. Details about the

dataset are described in the following paragraphs with

emphasizing major properties.:

 Flight Dataset: This dataset contains a

substantial amount of data, with around 30K

rows, related to flight operations over a month.

This dataset includes detailed information

about flight properties, aircraft features, arrival

and departure times, airport codes, flight status,

and various timestamps. The large volume of

data enables extensive analysis of flight

performance, on-time operations, and

airport/airline efficiency. It can support

decision-making processes, identify trends, and

detect anomalies in the complex world of flight

management.

 Baggage Dataset: This comprehensive dataset

contains detailed information about baggage

handling and tracking within an airport or

airline system. With approximately 1M rows, it

provides a complete record of baggage events,

statuses, arrival details, security-related

information, passenger profiles, and

timestamps for various handling stages

throughout the entire month. The scale of this

dataset allows for in-depth analysis of baggage

management processes, identification of trends

and patterns, and optimization of operational

efficiency across the entire baggage handling

system.

Major features regarding flight and baggage

dataset are summarized in Table 1 with feature name

and feature description pair details. These features are

interpreted in the upcoming sections.

Table 1: Major features of dataset.

Feature Name Feature Description

ACTUAL_TIME Actual time of flight.

BAG_NUMBER Number of baggage in

a single baggage

record.

BAGGAGE_EVENT Information about

baggage operation

itself.

BAGGAGE

STATUS

ALL Status of ba

BIM_CREATE_DATE

First timestamp when

baggage is entered to

the system.

BIM_ID ID of a baggage

record.

CATEGORY Identifies flight type

(Domestic,

International).

ESTIMATED_TIME Estimated time of

fli

FLIGHT_CODE Combination of

Airline Code and

Fli

ht Number.

FLIGHT

STATUS Status of fli

ht.

ID Flight ID (also

connection with Flight

Dataset).

SCHEDULED_TIME Scheduled time of

fli

ht.

SYSTEM_AIRPORT Identify airport from

the point of flight

direction view.

3.2 Pre-Processing

Raw data is transformed into clean data via several

pre-processing steps. These steps are divided into

three major parts as data pre-processing, outlier

analysis and repetition check. Data pre-processing

operations are done through data simplification and

data compression phases.

For each baggage data, feature column occupancy

rate is calculated. Columns with occupancy rate

below 100% are identified. . Columns with empty cell

are joined to achieve 100% occupancy rate.

 ATD (Actual Time of Departure) and ATA

(Actual Time of Arrival) are combined into

ACTUAL_TIME; STD (Scheduled Time of

Departure) and STA (Scheduled Time of

Arrival) are combined into

SCHEDULED_TIME; ETD (Estimated Time

of Departure) and ETA (Estimated Time of

Arrival) are combined into

ESTIMATED_TIME regarding arrival or

departure flight.

 SEAT_NUMBER, SECURITY_NUMBER,

SEQUENCE_NUMBER, FILE_NAME is

combined into PASSENGER_ID. i.e., 04B-33-

33.

 BAG_NUMBER with empty cell is filled with

the expected average bag number.

 The rest of the columns with empty cell are

excluded.

After completing pre-processing steps, the

following data analysis is conducted to help with the

rest of the pre-processing steps. For each flight code

with departure status, hourly baggage arrival pattern

before flight time is extracted in Table 2. For

example, flight code 1 is repeated 13 times for one

month period and 100 baggage has arrived when there

is more than 1 hour and less than 2 hours before flight

time.

Temporal Pattern Analysis of Baggage Impact on Flight Operations

833

Table 2: Hourly Baggage Arrival Pattern per Flight Code.

Flight

Code

Total

Baggage

Count

Flight

Count

Hourly Flight

Pattern (1 Hour, 2

Hours, 3 Hours, ...

)

1 308 13 [0,100,0, …]

2656

22 [0,2,867, …]

3 5325 23 [0,27,33, …]

For each flight code, average baggage arrival

versus time before flight is plotted as in Figure 1. Each

color represents a distinct flight code. Excessive

baggage count is observed as 1750. Also, it is found

that there can be baggage arrivals even 12 hours before

the flight. In this initial representation before cleaning

and pre-processing steps, there are many

outliers/anomalies regarding arrival pattern of baggage

records in Figure 1. Many baggage records seem

processed before too early hours than flight time.

Figure 1: Hourly Baggage Arrival Pattern per Flight Code

(Initial).

Outlier baggage records are excluded based on

rules such as empty PASSENGER_ID,

BIM_CREATE_DATE > ACTUAL_TIME and

earlier timestamps do not exist in pre-determined date

range. Repetition records are eliminated based on

evaluating cases of BIM_CREATE_DATE, BIM_ID,

FILE_NAME, ID, STATUS_INDICATOR and

PASSENGER_ID feature relations. After pre-

processing, hourly baggage arrival patterns for each

flight code are again visualized in Figure 2.

Furthermore, arrival pattern of baggage records is

distributed in a reasonable way. Figure 2 indicates

most activity concentrated in the few hours leading

up to the flight time. Finally, because of data cleaning

process, feature selection and feature merge,

dimensionality reduction is also obtained like from

(858K, 82) to (685K, 23).

Figure 2: Hourly Baggage Arrival Pattern per Flight Code

(Final).

3.3 Storage Strategy

Data can be classified as structured or unstructured.

Structured data fits neatly into tables, while

unstructured data cannot be easily mapped. Traditional

Relational Database Management Systems (RDBMS)

excel at handling structured data, but struggle with

large volumes of structured or unstructured data. To

address this, NoSQL databases have emerged as a

more scalable, flexible, and distributed solution

capable of efficiently managing both structured and

unstructured big data. These advantages make NoSQL

systems preferable for big data projects compared to

traditional RDBMS technologies.

This study area deals with big data regarding

millions of flights and baggage data with also future

projection. Additionally, it requires to be expanded

with real-time updates. Thousands of new rows will

be added daily from the airport system with each

flight, and new columns may be added as well. This

will result in a high volume of data and transactions.

Given these requirements, a NoSQL data

management system is the most suitable solution for

this big data study.

There are many NoSQL database technologies

used in the software field. After careful research and

requirement analysis process, it is decided that

Cassandra will be the best option for this study (Khan

et al., 2023). It is a wide-column store database,

which is a type of NoSQL database that can be used

for relational types of big data. Cassandra is an open-

source database that can be distributed across

multiple machines. It also provides a scalable,

maintainable system with high performance and fault

tolerance.

ICPRAM 2025 - 14th International Conference on Pattern Recognition Applications and Methods

834

3.4 Visualization

To be able to interpret the dataset in a better way, data

visualization is one of the most powerful methods

used in practice. Regarding this, some major

components of baggage dataset are visualized.

Distribution of flight categories can be seen in

Figure 3. There are 2 categories here:

INTERNATONAL and DOMESTIC. Their

percentage are quite close. Count of

INTERNATONAL flights is a bit more than

DOMESTIC flights in that snapshot.

Figure 3: Distribution of Flight Categories.

Distribution of flight status can be seen in Figure 4.

There are 5 categories here: ARRIVED, DEPARTED,

OPEN, CANCELED, CLOSED. Majority of the

flights belong to DEPARTED category. There are also

Figure 4: Distribution of Flight Status.

numerous ARRIVED and OPEN flights. However,

CANCELED and CLOSED flights are in negligible

range. Focus on the study is on DEPARTED flights

from the point of baggage operations.

The distribution of system airports can be seen in

Figure 5. There are 13 different system airport

categories. The top 5 system airports look like ALA,

ESB, ADB, ECN and TBS. ADB system airport is

going to be focused primarily regarding request of

data provider.

Figure 5: Distribution of Flight System Airports.

4 PATTERN ANALYSIS

Regarding pattern analysis of baggage records, two

major approaches are detailed here. Firstly, average

hourly baggage arrival pattern for each flight code is

calculated for several flights in distinct days. Pattern

Dissimilarity (PD) for a single flight here is defined

as shown in formula 1. Numerator b

is the absolute

difference for flight i at hour j and equals to

∣Hourly Baggage for Flight i at Hour j−Mean Hourly

Baggage for Flight Code at Hour j∣. Denominator a

is the mean hourly baggage for the flight code at hour

j. Please note that hour j here refers to delta before

flight time.

























)

Temporal Pattern Analysis of Baggage Impact on Flight Operations

835

Let us there are 3 distinct flights for a flight code.

Baggage count for 1 hour and 2 hours before flight

time are examined. Flight 1: 10 baggage at hour 1, 30

baggage at hour 2; Flight 2: 5 baggage at hour 1, 20

baggage at hour 2; Flight 3: 15 baggage at hour 1, 10

baggage at hour 2. Mean baggage count: (10 + 5 + 15)

/ 3 at hour 1 equals to 10; (30 + 20 + 10) at hour 2

equals to 20. PD for Flight 1 is calculated as (|10 – 10|

+ |30 -20|) / (10 + 20) and equals to 0.33 (33%).

Here, the measure of PD is needed to demonstrate

correlation with flight delay possibility. If there is an

increase in PD of a flight, it directly impacts flight

delay. For each calculated PD information, relation

between flights is analyzed based on a threshold

approach. A flight is marked delayed if the time

difference between ACTUAL_TIME and

SCHEDULTED_TIME is greater than 15 minutes.

Overall number of delayed flights are counted and the

delay ratio for each threshold is calculated. So that the

correlation between PD ratio and flight delay is

shown in Table 3.

Table 3: Correlation Between Pattern Dissimilarity Ratio

and Flight Delay.

Pattern

Dissimilarity

Ratio

Threshol

Flight

Delay

Ratio

Delayed

Flight

Count

Total

Flight

Count

30 57.30% 2199 3838

40 60.36% 1457 2414

50 61.75% 980 1587

60 66.98% 720 1075

70 68.61% 518 755

80 69.96% 396 566

90 71.87% 327 455

100 73.68% 266 361

Flight delay ratio over PD ratio threshold in given

sample data in Table 3 is displayed in Figure 6.

Results indicate that percentage of unexpected

baggage records affect flight operations negatively

such as delay in our case.

Figure 6: Pattern Dissimilarity Ratio – Flight Delay

Correlation.

As a second pattern analysis of baggage records,

a new feature is generated with name of

MULTI_BIM_CREATE. It is calculated for each

baggage record as the number of repetition baggage

records with same ID, BIM_CREATE_DATE,

PASSENGER_ID having STATUS_INDICATOR

changed or deleted. Repetition records may appear in

baggage dataset due to several reasons such as

baggage rehandling, integration from multiple

sources. For each flight, the percentage of baggage

records that has repetition (MULTI_BIM_CREATE

is greater than 1) over total baggage records is

computed as shown in Table 4. For example, Flight

ID 4 has total 154 baggage records. 42 records have

repetition. So, MULTI_BIM_CREATE_RATIO is

calculated as 27.27%.

Table 4: Multi BIM Create Ratio per Flight ID.

ID MULTI_BIM_CREATE_RATIO

1 0.5 (1/199)

2 3.33 (2/60)

3 13.7 (10/73)

4 27.27 (42/154)

Correlation between MULTI_BIM

CREATE_RATIO and flight delays is displayed in

Table 5. A flight is marked as delayed if the time

difference between ACTUAL_TIME and

SCHEDULTED_TIME is greater than 15 minutes.

For example, there are 464 flight records that have

MULTI_BIM_CREATE_RATIO >= 15% and

66.16% of flights have delay in the first row.

Table 5: Correlation Between Multi BIM Create Ratio and

Flight Delay.

MULTI_BIM_C

REATE_RATIO

Delay

Ratio

Delayed

Flight

Count

Total

Flight

Count

15 66.16% 307 464

16 66.83% 268 401

17 69.01% 236 342

18 70.3% 213 303

… … … …

59 90.0% 9 10

60 100.0% 7 7

Flight delay percentage over MULTI_BIM

CREATE_RATIO threshold is displayed in Figure 7.

It indicates that percentage of repetitive baggage

records affect flight operations negatively such as

delay.

ICPRAM 2025 - 14th International Conference on Pattern Recognition Applications and Methods

836

Figure 7: MULTI_BIM CREATE_RATIO – Flight Delay

Correlation.

5 CONCLUSIONS

Maintaining baggage requests of passengers in an

airport may have direct or indirect effect on other

operations such as ground services, flight. In the

study, temporal pattern analysis of baggage

operations is investigated to determine if there is a

correlation for flight delays in a holistic perspective.

Several pre-processing and cleaning strategies are

applied on the given dataset with also considering

cross relation between baggage features and flight

features. Gathered results reveal two major founding.

Firstly, creating multiple baggage records per

passenger has a negative impact on the related

departed flight operation. Secondly, increase in

pattern dissimilarity ratio for baggage arrival

correlates with flight delay possibility.

In future, extended version of dataset is going to

be analyzed with the current systematic approaches

and founding. Only statistical analysis may not be

sufficient or flexible enough to manage the growing

volume of data and the increasing number of features.

To address this, the focus will shift towards

incorporating AI-powered solutions to enhance the

understanding of the effects of baggage records and

operations on flight delays.

Developing AI-driven models can accurately

predict baggage counts with daily, hourly, and even

minute-level precision, considering both airport-

specific and flight-specific factors. For this purpose,

machine learning (Random Forest, Gradient-

Boosting etc.), deep learning (Neural Network, CNN,

LSTM etc.) and time-series (ARIMA, SARIMA etc.)

approaches are planned to be utilized. This will

enable airport operators to proactively manage

baggage handling resources and optimize their

operations, reducing the impact of baggage-related

issues on flight delays.

ACKNOWLEDGEMENTS

This study was supported by Eureka-ITEA Project

"SOCFAI" (Project Number: ITEA-21020). We

extend our gratitude to TUBITAK for funding this

project. Our special thanks go to our project partners,

TAV Technologies, Siemens A.S for their invaluable

contributions and collaboration. Additionally, thanks

to SOCFAI Project Team for their technical

contributions during the initial phase of the project.

REFERENCES

SOCFAI. (2024, October 15). Secure Open Collaboration

Framework powered by Artificial Intelligence

Homepage. https://www.socfai.com/

Anupkumar, A. (2023). Investigating the Costs and

Economic Impact Of Flight Delays In The Aviation

Industry and The Potential Strategies for Reduction.

Amiri, M. A., & Kusakci, A. O. (2024). A Scoping Review

of Artificial Intelligence Applications in Airports.

Comput. Res. Prog. Appl. Sci. Eng. Trans. Ind. Eng, 10,

2900.

Sayın, M. G., Aktaş, D. Y., Bolat, M., Çelenli, M. K.,

Dursun, B., Koç, G., & Üçkardeş, K. S. (2023). A Study

Of Predicting Arrival Patterns Of Airport Passengers

To The Counters On The Basis Of International

Terminal. Avrupa Bilim ve Teknoloji Dergisi, (51), 63-

74.

Chen, S., Park, C., Guo, Q., & Sun, Y. (2024). Advancing

a major US airline’s practice in flight-level checked

baggage prediction. Intelligent Transportation

Infrastructure, 3, liae001.

Jiang, B., Ding, G., Fu, J., Zhang, J., & Zhang, Y. (2024).

An Overview of Demand Analysis and Forecasting

Algorithms for the Flow of Checked Baggage among

Departing Passengers. Algorithms, 17(5), 173.

Ochoa Barnuevo, M. L. (2023). Enhancing Baggage

Handling Duration Predictions for KLM: A Data-

driven and Machine Learning Approach Using Camera

and Sensor Data (Bachelor's thesis, University of

Twente).

Guo, X., Grushka-Cockayne, Y., & De Reyck, B. (2020).

London heathrow airport uses real-time analytics for

improving operations. INFORMS Journal on Applied

Analytics, 50(5), 325-339.

Le, V. T., Zhang, J., Johnstone, M., Nahavandi, S., &

Creighton, D. (2012, October). A generalised data

analysis approach for baggage handling systems

simulation. In 2012 IEEE International Conference on

Systems, Man, and Cybernetics (SMC) (pp. 1681-

1687). IEEE.

Temporal Pattern Analysis of Baggage Impact on Flight Operations

837

Petersen, P., Stage, H., Reis, P., Rauch, J., & Sax, E. (2024).

Comparison of Dimension Reduction Methods for

Multivariate Time Series Pattern Recognition.

In ICPRAM (pp. 809-816).

Khan, W., Kumar, T., Zhang, C., Raj, K., Roy, A. M., &

Luo, B. (2023). SQL and NoSQL database software

architecture performance analysis and assessments—a

systematic literature review. Big Data and Cognitive

Computing, 7(2), 97.

ICPRAM 2025 - 14th International Conference on Pattern Recognition Applications and Methods

838